将via标注的内容提取成yolo格式

ni哥跑的快

已于 2023-05-11 15:07:50 修改

阅读量377

点赞数

文章标签： YOLO json python

于 2023-05-03 16:31:36 首次发布

本文链接：https://blog.csdn.net/Apple0631/article/details/130474328

版权

该文描述了作者在创建AVA数据集时遇到的问题，即Yolo对目标检测的效果不佳，影响了在Slowfast中的回归效果。作者决定使用VIA工具标注的数据来训练Yolo。文中提供的代码展示了如何处理标注文件，将json格式的标注信息转换为适合Yolo训练的txt格式，包括处理不同类型的标注方式。

摘要由CSDN通过智能技术生成

本人在生成ava数据集的时候发现yolo对目标的检测情况不大好，导致最终放到slowfast中回归效果差，因此决定将via标注好的数据放到yolo中进行训练。

我大概解释以下，在标注的文件中一个json对应了一个视频的多个图片，这里拿我编辑的视频1解释：image1是视频1里面第一张照片，后面的_1是里面有一个对象。这里就是第一个视频的第一张图片有一个检测对象。

        "image1_1": {
            "vid": "1",
            "xy": [
                2,
                0.0006400000000184036,
                6.999840000000006,
                1245.99936,
                713.00016
            ],
            "av": {
                "1": "0",
                "2": "",
                "3": "",
                "4": "",
                "5": ""
            },
            "flg": 0,
            "z": []
        },
        "image2_1": {
            "vid": "2",
            "xy": [
                2,
                1,
                3.26914,
                444.11931,
                171.65494
            ],
            "av": {
                "1": "0",
                "2": "",
                "3": "",
                "4": "",
                "5": ""
            },
            "flg": 0,
            "z": []
        },

如果在用via标注的过程中回来标注前面的图片，则会变成这样：

大概意思就是我分别在第12张照片和第13张照片中补充了一个对象的检测框

"12_HndcTSRv": {
            "vid": "12",
            "flg": 0,
            "z": [],
            "xy": [
                2,
                437.31,
                6.247,
                441.996,
                82.777
            ],
            "av": {
                "1": "0",
                "2": "",
                "3": "",
                "4": "",
                "5": ""
            }
        },
        "13_4XM1e1nr": {
            "vid": "13",
            "flg": 0,
            "z": [],
            "xy": [
                2,
                546.638,
                4.685,
                452.928,
                106.204
            ],
            "av": {
                "1": "0",
                "2": "",
                "3": "",
                "4": "",
                "5": ""
            }
        }

介于我标注的结果有以上两种情况，我撰写以下代码解决问题

以下是转化代码：

import os
import json
from PIL import Image
import re

def extract_key_value(json_file, target_key):
    with open(json_file, 'r',encoding='utf-8') as file:
        data = json.load(file)
    
    if target_key in data:
        return data[target_key]
    else:
        return None

def main():
    folder_path = 'C:/Users/yuqi zhang/Desktop/finish2'
    img_path = 'C:/Users/yuqi zhang/Desktop/finish2/img/'
    target_key = "metadata"  # Change this to the desired key
    save_path = 'C:/Users/yuqi zhang/Desktop/data'
    
    #打开文件夹读取json文件中value的内容
    for root, _, files in os.walk(folder_path):#打开文件夹
        json_files = [f for f in files if f.endswith('.json')]
        for json_file in json_files: #打开文件夹的文件
            # with open(json_file, 'w') as file:
            file_seq=json_file.split("_")[0]
            # print(file_seq)
            json_path = os.path.join(root, json_file)
            # print(json_path)
            value = extract_key_value(json_path, target_key)
            # print(value)

            #构建对应的txt文件
            for i in range(1,17):
                img= 30+i*15 +1
                img_seq=str(img).rjust(6, "0")
                txt_file=file_seq + "_"+img_seq+".txt"
                
                # regex_pattern = str(i)+"_"
                print("-------------------------------------------")
                contentList1 = search_json(value, "image"+str(i)+"_")
                contentList = search_json(value, "^"+str(i)+"_")
                
                txt_path = os.path.join(save_path, txt_file)
                with open(txt_path,"w") as txt_file:
                    # print(results)
                    # print(len(contentList))
                    for i in range(len(contentList1)):
                        x=contentList1[i][1]
                        y=contentList1[i][2]
                        w=contentList1[i][3]
                        h=contentList1[i][4]
                        print(x,y,w,h)
                        if x < 0 : x=0
                        if y < 0 : y=0
                        if w < 0 : w=0
                        if h < 0 : h=0

                        img_file=file_seq + "_"+img_seq+".jpg"
                        img_p=img_path+img_file
                        img = Image.open(img_p)
                        img_w=img.width
                        img_h=img.height

                        x1=round((x+w/2)/img_w,6)
                        y1=round((y+h/2)/img_h,6)
                        w1=round(w/img_w,6)
                        h1=round(h/img_h,6)

                        txt_file.write(f"0")
                        txt_file.write(f" {x1}")
                        txt_file.write(f" {y1}")
                        txt_file.write(f" {w1}")
                        txt_file.write(f" {h1}\n")




                    for i in range(len(contentList)):
                        x=contentList[i][1]
                        y=contentList[i][2]
                        w=contentList[i][3]
                        h=contentList[i][4]
                        print(x,y,w,h)
                        if x < 0 : x=0
                        if y < 0 : y=0
                        if w < 0 : w=0
                        if h < 0 : h=0

                        img_file=file_seq + "_"+img_seq+".jpg"
                        img_p=img_path+img_file
                        img = Image.open(img_p)
                        img_w=img.width
                        img_h=img.height

                        x1=round((x+w/2)/img_w,6)
                        y1=round((y+h/2)/img_h,6)
                        w1=round(w/img_w,6)
                        h1=round(h/img_h,6)

                        txt_file.write(f"0")
                        txt_file.write(f" {x1}")
                        txt_file.write(f" {y1}")
                        txt_file.write(f" {w1}")
                        txt_file.write(f" {h1}\n")


def search_json(data, regex):
    result = {}

    contentList = []

    for key, value in data.items():
        if re.search(regex, key):
            result[key] = value['xy']
            contentList.append(value['xy'])

        elif isinstance(value, dict):
            sub_result = search_json(value, regex)
            if sub_result:
                result[key] = sub_result
    # print(result[key])

    return contentList

if __name__ == '__main__':
    main()